← Back to search
Episode 24: We Have Zero Loyalty to AI Labs
Fired By Robots · 2026-11-18 · 70 min
Show full episode description
A few episodes ago I dropped Claude Code and went all in on OpenAI. Now it's the opposite: ever since Opus 5.5 it's been Anthropic, Anthropic and Anthropic. Is that a problem? We don't think so. There is no loyalty to AI labs, and there shouldn't be. We start with nostalgia. Aivars found an old video of himself writing code by hand, with winter outside the window, and it made us miss the slower days before AI did the debugging for us. Then Aivars shows why you still need to understand code: Claude tried to rewrite his whole paywall for a Black Friday offer that needed one if statement. Also in this episode: - Fable 5.2 and why the $200 plan finally feels comfortable - will AI create more app builders, and is SaaS dead? - GPT-6.2 Astra and the Anthropic IPO - using ChatGPT for computer use, and why Opus feels better to work with - how Claude writes the prompts for our thumbnails and OpenAI draws them - T3 Code vs the Claude app, plus model-agnostic tools like Hermes Agent and Grok Bot - a follow-up on the $100K hackathon Chapters:
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
Two indie developers weigh what AI coding took away from the craft, whether SaaS and consumer apps still have room, and why they switch between
Anthropic,
OpenAI and tools like T3 Code with zero loyalty.
Benefits
- Fable 5.2 fits comfortably within a $200 plan at ~60-70% weekly usage
- Claude orchestrating Codex code review inside one pull-request routine
- Claude prompting OpenAI image models yields better thumbnails than direct prompting
- T3 Code gives one model-agnostic interface across Mac and Linux via Tailscale
- Switching labs freely keeps you on the best model for the job
Use cases
- Black Friday one-time-purchase paywall for lapsed subscribers: Claude over-engineered what should be one if-case
- ChatGPT reconciles monthly bank statements against receipts and pulls missing invoices from email
- ChatGPT computer use: ~50 photos of ~10 items enhanced and listed on a Polish resale platform automatically
- Opus 5.5 edits podcast episodes and prompts OpenAI to generate each episode thumbnail
- Claude cron job writes an SEO article every three days for one of Paweł's websites
KPIs / results
- Fable 5.2 usage ~60-70% per week on the $200 plan; early Fable burned limits in hours
- Mark Lou's $100K hackathon: $99 entry fee dropped to free after backlash; ~3,000 sign-ups
- Aivars signed up for a 21 km half marathon in late January
Tools / build
- Fable 5.2 as project manager delegating to Opus and Sonnet
- Claude + Codex pull-request review routine
- T3 Code multi-machine setup over Tailscale
- Grokbot virtual assistant with X integration
- Custom mobile app for managing Hermes cron jobs
- ChatGPT tax and bookkeeping document checker
📑 Chapters — tap a time to jump there
00:50
Welcome
- Paweł and Aivars (Latvia, living in Thailand) kick off Fired By Robots
- Outages this week; enough news to record daily
01:58
Missing hand-written code
- Aivars found old footage of hand-coding SwiftUI by a winter window and felt nostalgic
- Debugging was the first thing AI took; the one-line-fix satisfaction is gone
- Building is now an endless treadmill: build, release, repeat with no downtime
09:25
Outages: back to hand-coding?
- Claude outage last week prompted 'have we been in a fever dream?' tweet
- Paweł's first app took a year and flopped; marketing and ASO mattered more than code
11:20
Woodworking and hardware
- Woodworking, Raspberry Pi tinkering and AI-generated 3D prints as possible new crafts
12:20
Fable 5.2 and usage limits
- Aivars uses Fable 5.2 as project manager, delegating subtasks to Opus or Sonnet
- Less token-hungry: ~60-70% weekly usage on the $200 plan vs limits gone in hours at launch
15:13
Model memory and over-engineering
- Claude's memory helps context but anchors brainstorming to old beliefs
- Black Friday one-time-purchase offer: Claude rewrote the whole paywall instead of one if-case
- Non-technical founders risk living with over-engineered spaghetti legacy
18:40
Will AI create more builders?
- AI adds a few more builders, mostly marketers in mobile, but everyone building means fewer customers
- 2024-25 was a short window where AI-enabled builders could make money
21:28
Is SaaS dead?
- SMBs with any dev capacity can build their own tools; Marc Lou's email service example
- SaaS shifting toward 'software as part of a service'
22:54
Is there room for consumer apps?
- ChatGPT threads mixing protein tracking, half-marathon training and code get messy
- Single-utility apps (21k running coach, calorie counter) still have demand
- Long term, big AI and apps inside ChatGPT eat more consumer needs
29:11
GPT-6.2 Astra and the Anthropic IPO
- GPT-6.2 Astra got harsh reviews; seen as a jab timed to the Anthropic IPO
- Paweł barely used OpenAI models for two months since Opus 5.5; may drop to $20 plan
32:27
ChatGPT for computer use
- ChatGPT/Codex app is better at computer use: tax receipts matched to bank statements, files sorted
- ~50 photos of ~10 items enhanced and listed on a Polish resale platform automatically
- Fable implements, then asks Codex to review the pull request
37:28
Why Opus feels better
- Opus is pleasant to work with: no em-dash fluff, understands you — not visible on benchmarks
- Sol 5.6 was loved in May-June; initial Fable smarter but less pleasant
39:55
No loyalty to AI labs
- Paweł once dropped Claude Code for OpenAI; now the reverse
- Both models can build nearly anything; Claude has better design taste
41:12
Thumbnails with Claude and OpenAI
- OpenAI image generation beats Claude for thumbnails, though both are mediocre
- Opus 5.5 edits the episode and prompts OpenAI to generate the thumbnail — better than direct prompting
- Pay for whatever model is most cost-effective; no loyalty to corporations
45:04
Trying T3 Code
- T3 Code: abstraction layer that assigns subtasks across Claude, Codex, Grok
- Aivars set it up for Android-on-PC and iOS-on-Mac but questions why he needs it
- Claude app on both machines over Tailscale already covers multi-machine tasks
48:40
Model-agnostic tools
- Model-agnostic tools like T3 Code fit a no-loyalty world; best mobile experience
- Grokbot best for virtual-assistant tasks and X integration; Hermes needs tinkering
- Paweł built his own mobile app to manage Hermes cron jobs but dislikes maintaining it
52:11
T3 Code vs the Claude app
- Paweł keeps T3 Code for coding, Grokbot for assistant work; likes the separation
- Claude cron job writes an SEO article every three days; T3 mobile app manages remote Mac/Linux tasks
- Aivars does analytics via Claude pulling App Store Connect data, then unsure where to implement
59:57
Building a custom assistant
- Paweł considers replacing the Grokbot subscription with a wrapper on Hermes, Claude or Grok
- Tinkering on a custom assistant won't bring clients — marketing will
61:30
The $100K hackathon follow-up New episode every week – subscribe so you don't miss the next one.
- Marc Lou's $100K Higgsfield hackathon dropped the $99 entry fee after X backlash; now free
- Started Monday the 16th, 30 days, ~3,000 sign-ups; neither host is entering
- Revenue-in-30-days rules disadvantage subscription apps
OpenAI, Anthropic, OpenAI, Anthropic, but ever since 5.5 it's been on Tropic, Anthropic and Anthropic. I remember that few, quite a few episodes ago you unsubscribed from cloud code and basically doubled down on OpenAI and now it's opposite. Yeah, yeah, exactly. So it's kind of hard to be in that like, in this kind of business because there is no loyalty in my opinion for the models. Hello, welcome to another episode of Fired By Robots podcast. My name is Paweł Karniej and as always with me is my friend from Latvia that lives in Thailand right now. Hello, Aivars. How are you doing today? Hi Paweł, doing great. Thank you. How about you? I'm doing good myself. I am not a potato. I kind of slept six hours, which is good enough to feel decent. But yeah, it's great. We have some topics today to talk about. We have some stuff that I saw you posting about on social media. There was some outages happening this week. And yeah, it's kind of like interesting week as always. We talked about it last week that we could basically record a full episode every day. And there are all the time news to cover like in our technology space. But I wanted to start with this one because I saw you kind of have this video. I don't know if it went viral, but you had some amounts of likes and replies on X that you have nostalgia for the good old days when we were writing code by hand. Hmm. Yeah. All right. Yeah. I posted that one. I somehow were looking through basically, I was cleaning storage as always on my computer. And I found basically, I have some video files basically locally. And that was from like maybe two, maybe even three years ago. Just footage of me kind of sitting and programming, typing code by hand in front of the window with winter outside. And yeah, I don't know. It kind of, I was watching that video and I was thinking, yeah, you know, I'm kind of missing good old days, you know, when you were able to slow down and basically sit and work on one task and write code and compile and check the results or previews in SwiftUI kind of previews. And I don't know. Now it's gone. Now it's gone. Now it's gone. Yeah. I always loved this kind of vlog types, videos when like people sit down and cold. And I like I did one or two attempts at recording this kind of videos in the past. But I always loved your video because you have like a good feeling about like recording those and your videos always look great. And yeah, it kind of made me feel a little bit nostalgic too. So it's kind of nice. And I don't, I wouldn't say I miss the times, but it kind of like the feeling of this kind of accomplishment. Like whenever you figure out something, like you spend, like, I think debugging was one of the first things that AI took from us. And, and I remember, I remember that was like the feeling you had, like when you're trying to figure out something without AI, you know, and even like if it, if it take, it took you like a whole day to figure out, okay, that was the issue. And usually the bugs that caused you like whole day of, of issue that you were going through cold. It was almost always one line change in cold or like missing semi colon or something like that. And, and, and, and I, I feel, I feel like this kind of accomplishment feeling of solving that bug was satisfying part of our job that don't, no longer exist. Just, uh, I think, uh, when I think, uh, when I look back at my beginning of the, of the year, um, like, uh, I, I, I, I was full on in AI psychosis. I tried to build, I don't know how many apps. I don't know. I started 100 projects and now I look back when I took a little break in September and October from building stuff. I look back, it was kind of a craze. But right now, it's like we spoke about it. And it's way more important to think, okay, what to actually build right now whenever you can build anything. But yeah, those were the days, you know? Yeah. Yeah, previously it was like, you know, you're coming up with some idea or like theory that you would like to test. And then you're basically spending hours or even days kind of building that writing code, you know? Currently, it's like you come up with a day and it's basically like that. You can create it in a moment or if that's something for real bigger like application from the scratch, okay, that will take a little bit more. But anyways, that can be built in like no time. Yeah, that's kind of cool. But also it's like endless kind of running on this treadmill, you know, kind of, okay, I heard that today. I can build it. I'm releasing it. Now I'm kind of waiting for results, what I'm doing next, you know? And there is no kind of downtime. There's no like chill time, you know? Previously, you were able to spend a week, you know, kind of working on something, building on something. Then you release it, then you're waiting for results. I think it, yeah, sure, it was way slower and way more inefficient way to build software. But also there were kind of way less stress in the process, you know? It was way more chill and way more laid back. So probably that's what I'm missing. I don't know. But yeah, I was looking into that kind of, in that video and I was kind of thinking, yeah, that was cool actually. Yeah, that was cool. I don't know what is next cool thing, you know? Because kind of this building stuff and playing with AI models and catching up on new models, testing them, that's kind of interesting, but I don't have that feeling that I had before when I was kind of, let's say, crafting applications. And yeah, you can say, yeah, you can just still keep writing code by hand. Yeah, sure, you can do that. You can also do whatever as a hobby, you know? Woodwork, what else not, you know? It's, yeah. Previously it was like creative, but job and process. Now it feels like waste of time when you're writing code by hand. Yeah, yeah. For me, like, I was like kind of browsing Twitch a little bit, like, I don't know. I think after our podcast, I was trying, last week recording, I was looking to like, just to play something. And I saw some girl, I don't remember her name, but she was kind of doing programming by hand, which was, which, which was, which kind of was weird for me. Because I don't remember when I was programming by hand. I don't remember when I did this. And for sure that has an impact on like my ability to code. Of course. The, the skill is no longer, like the coding skill is no longer needed. Although, we don't know what's going to happen this week. We've had so many like outages and everybody's scaring us. Does it happen? Well, does, do we, do we go back to like hand coding? Because like, maybe there, the, the, the data centers will get hacked and there will be no models running. We had, like, I think we had some cloud usage, no, no usage, cloud outage. I think that last week that happened. And like, I wrote a tweet that some people replied to like, have we been in a fever dream that we had AI build software for us? I don't know. Sometimes I feel like, I would like, I would like to go back to that times because right now I, I know way more about what's actually really necessary instead of like, when I like to make the app successful. And that's marketing, not coding actually. But I remember like, my first app, which I released, took me like, I don't know, two years to release or, or a year, I think a year. And that flopped spectacularly because I didn't know, back then, I didn't even know anything about ASO at all. So, yeah, it's been good times. And I feel maybe woodworking, that's something cool still. I was kind of thinking for a while, oh, it would be cool to be, to get into some small hardware electronics making, you know. I don't know, tinkering with just Raspberry Pi and there is like, you can use AI to generate you 3D prints using 3D printers and make all that cool stuff. I think that's, that's kind of cool things. I see people are doing, but, yeah, it requires you to, like, I think there is some kind of cool stuff doing, but will I move to doing that? I'm not sure. I'm not sure. Yeah. Yeah. So, yeah, in the past week, we've had Fable 5.2 release, and I've been, I like, last week, we talked about it. I, like, I tested it a little bit, but I, like, I don't know if my, my tasks are so trivial right now that basically Opus can handle all of it. I'm not sure, but have you tested the Fable 5.2? Oh, yes, yes, I'm using it. my usage of models is basically I'm, lately I'm just going with Fable for, like, general, I'm just giving tasks to Fable and asking to open, like, subtasks with whatever model Fable will decide to use for that specific task, and sometimes that's Opus doing programming, sometimes that's Sonnet, but Fable is basically kind of the, like, let's say, project manager on that specific thing that I would like to do, and I'm using Fable 5.2 now. I don't know if it's smarter than 5.1, but probably less resource demanding, less token hungry, because I see that I basically can comfortably survive on my $200 plan, which is ultra now probably something changing names, yeah. Yeah. I see from my usage that the cheaper plan will not be enough because I'm kind of hitting more than 50% in a week, something like 60, 70, something like that. so it's comfortable place to be with. I see that I have a little bit more capacity if I will work harder any week, but I definitely not hitting limits now. So yeah, comparing with what was happening when Fable just was released first time, it was like few days of usage and all the limit of Fable are gone. Few hours, yeah, I agree. Depends on task, but basically you're right, you were able to burn it in a few hours. I was kind of trying to stretch it as long as possible, but it was still maximum few days, you know, but currently with Fable 5.2 it's quite comfortable. And yeah, its model feels smart, sometimes you're getting that AI dumbness, you know, and just taking basically what I like and don't like about models. I like that this cloud, basically memory, it builds up and it becomes smarter and it's easier to talk because that model understands context and you don't need every time to basically give your intro or what else not. But that's also a problem because when you're trying to look for some new ideas, new solutions or brainstorm something new, it kind of sticks with old beliefs and old concepts that we already did a week ago and basically keeping that like whatever, a rule or a fact which is not changeable, you know, and so on. And in programming as well, I still see that it's, yeah, we can, you're right, we can skip the reading of the code, but it's still keen to and try to over engineer solutions sometimes, you know. I can give you a more specific example. For example, I was working on Black Friday offer and I was kind of thinking, okay, it's Black Friday, what if I will offer like one-time purchase for my customers who never converted on like regular subscriptions, like a Black Friday deal. Okay, I understand you don't like subscriptions, I prefer subscriptions as a developer, you as a customer don't like, but maybe you want to have pro version, I will give you like one-time purchase. And I just explained the logic and what I would like to complete and cloud just started to rewrite all the kind of paywall functionality and add extra logic and so on, but that basically should be just one if case, you know. So, if it is just new customer downloading application, show regular paywall. If it is customer who uses the application for two weeks or more, basically keep using it, didn't convert or converted, but then canceled subscription, show this offer. That's it, so simple, just one if case and one more kind of product in the app store and one more paywall. That's it. And I saw how it could go really wrong, you know, if I do not understand programming, if I just this non-technical founder, it probably would work, probably cloud will build something which will work and basically pass the user test, but it would be already so overcomplicated and you have to live with that legacy all the time after that and then you will build on top of this spaghetti, you know, your next kind of idea or next experiment. It can go wrong. So I do not regret that I spent all those years learning programming and I'm still not kind of over optimistic about the AI kind of ability to make right decisions. I think that writing code, yes, it's good. I can basically skip reading the code, but the general logic sometimes it makes dumb mistakes. I don't know. Yeah, yeah, I totally agree. Like, I think like, of course you can build this kind of small app without needing the technical knowledge and of course you can build even bigger apps right now, but I still feel like it won't increase the amount of people that will build apps by a huge amount. Yes, there will be a few more builders and I think in the mobile app space especially there will be more builders because it's a perfect space for people who know how to do marketing direct to consumer go-to and like previously if it held you back yes, you can build anything right now but everybody is building anything right now so there are no customers basically what I'm like I'm trying to say that we've had I think we have been like in the past in 24 25 we have been that short period of era where like whether you are thinker and build AI has enabled you you could make some money with doing stuff with AI but I'm not sure like that's not going to be the case long in my opinion like it enabled marketing teams it enabled people with like budgets that can like literally dump money into marketing an app or something like that basically most of people outside of our industry who were thinking about building some stuff they already started building this stuff some time ago and most people understandment of AI is still just a chatbot you know with bigger and bigger knowledge what I'm trying to say like before people will actually learn that there's stuff to build with AI there will no longer be stuff to build with AI that's what I'm trying to say because everything will be free available online or like eaten by big AI that's what I'm trying to say what do you think about this I don't know I think kind of building SaaS it because if software as a service is targeted to medium and small businesses then probably and if they have any development capacity they will probably can build that stuff by themselves kind of fine tune it for their own needs so therefore I think this kind of saying that SaaS is dead or dying there is some kind of truth in that because yeah sure we have those examples someone built like whatever you know Lightroom for example Adobe products which are kind of glitchy and doesn't really work but maybe later it will be possible to build them so Mark Lowe built emailing service to just replace more expensive kind of tool with cheaper one but probably spent more money on building it and kind of could pay in a few years for using existing services so it's very kind of tricky and you have to look case by case it's yeah I think it's more and more hard to build SaaS for businesses but do you consumer do I eat enough protein you know this kind of new thing nowadays everyone worry about protein and I'm just basically dropping my food pictures and explaining what I'm eating to chat bot that dots you know thing and just kind of asking it what it's good if I started to kind of running again and I signed up for half marathon at the end of January and to run half marathon which is not like super crazy distance it's 21 kilometers but you better prepare at least little bit you know so I started to run and also again I asked chat GPT for some kind of okay what will be the program for running and so on and it becomes already messy you know I'm talking about marathon in this I'm talking about food and basically then I'm throwing some programming tasks into the chat GPT and I feel that probably could be cool to download some 21k application or build one for myself just to do it exactly like that you know to have my plan what I'm running and basically feed it with my run results and basically see how it works do I need to is that I think that there are still so many users outside who are looking for single utility application which will basically fix the one problem like running 21k or counting calories or whatever flashlight being a flashlight up on your phone you know which is generally super simple application you like said like there is still space with consumer and definitely and there is still space with SaaS like I think it's kind of changing basically right now into towards something an actual service software as a service not the main service but you bring some software as part of your service that's kind of one angle and that's like more service business than it was before I totally agree there is and there will be time in consumer space for next few years to create like that AI creates some new problems that we want to solve with AI and I don't know what kind of problems are those like I had like for example I had an app idea for the longest time I that I just remembered about it I forgot about like when I was like creating using my app AI vidly which I sold last year and I was using my phone to generate content like AI images and videos I thought of an app idea to clean up space on my disk or actually sort out the AI generated items from the regular photos you know that's an app idea I thought of myself that I could build I never build it maybe I will at some point there are multiple app cleaners like there's there's a ton of things but like like like you said people will move more towards solving their issues with AI inside Charged GPT app and it will get better right now because I think like within Charged GPT or I don't know like I don't use the cloud app on too much but you can build basically apps inside Charged GPT right now to like to track your calories or something people are doing that on social media and everything is going towards this kind of situation where it's going to be cheaper and cheaper and at some point free but yes like I think in general the direction is like big AI will take care of most of consumer needs with some apps like in the longer perspective in the short perspective I believe what you said is actually true but in the longer perspective I think big AI is going to eat a little bit more yeah that's what I'm going to finish this total we start with Fable 5.2 I wanted to talk a little bit about Astra 6.2 I saw some not so kind words online for that and that's kind of interesting for me because I've been like watching Tio I mentioned Tio T3 GG a couple times on the podcast so I won't be like mentioning too many other people but like he basically said like Astra is really in it's good to work with and people say that like Astra 6.2 is a pretty good ad for Anthropics IPO that is happening today today something that we can talk about today or tomorrow I'm not 100% sure so I think they tried to release the model that could take a stab at Anthropic a little bit IPO that's happening and I think it backfired you know I haven't used it I actually haven't used too much of OpenAI models in the past like two months almost at all I like ever since Opus 5.5 came out it feels so good to work with that model it's like I don't even know like my work with Fable is basically the same as you do so I have opened my OpenAI subscription and I probably will going to lower my usage to $20 subscription just to keep like I still use ChatGPT app for some stuff but they are not doing well they are not doing low ever since that Opus 5.5 it's not been going well for OpenAI I have feeling that Codex application or ChatGPT application has little bit better better computer usage kind of skills that's why I like it and being like not employed person I have to do my taxes you know and yeah sure I have a content and not a crazy person but I at least few hours in a month at the end of the month I basically opened a bank statement and then I was looking okay do I have the ticket for each of business expense that I had or invoice or something currently I just kind of downloading that giving to chat GPT and saying okay please check if we're missing something and if we're missing something sometimes not sometimes but most often that is basically in my emails and I that you just need some simple AI model which is able to basically handle email and files on your Mac or PC and yeah it just downloads files put those in the right folders and that works nice I'm pretty sure cloud also can do that but I try to use both over time for like tasks like that you know to handle files and so on and somehow chat GPT was better but otherwise I'm quite I will say I'm in about the same shoes like you I'm mostly using the Fable for all the kind of creative work programming tasks and so on and then I'm using codex as like second pair of hours just to review the code and basically I'm not even kind of opening codex or something application I'm just asking Fable to have that as a routine you know to do the task implement everything and then ask codex for review and then they're basically working together on pull request and that's it you know yeah my usage of open AI models has been exactly the same I have some tasks that require computer use and that's what I use charge charge gpt app codex app or charge gpt app I don't even remember they they are consolidating them so yeah it's charge gpt application but you want to stay in codex tab and that's messy that's why I don't like this application yeah yeah my usage is pretty much the same that's what I use it's funny because I use for taxing stuff I use the open AI models I am doing some documentation through it right now I'm doing some like this for my tax and yeah I'm doing that with chat gpt because like the computer research is better I for example I did it like one or two months photos of items I want to sell on popular in Polish on Poland platform to sell used stuff basically and I gave like I don't know 50 photos maybe of like 10 items or something like that and it basically I gave it a task okay those are the photos maybe try to enhance it a little bit maybe make it look a little bit more presentable and then create offers on this platform use my computer and it did it automatically I was just like here in my office I just want making photos of the items then send it into the gpt thread and it was like doing and it kind of cool automation stuff I because it's important to have a model that's good to work with right like Opus is so good to work with and that's why I didn't care a little bit I care too much about the model needing to be way more intelligent I don't see too much difference in Fable being way more intelligent or maybe it is maybe I don't do any tasks that are so kind of out of my reach Opus is so good to work you talk to it it doesn't write to you like thousand M dashes it doesn't like fluff the output is that is not visible on the benchmarks that much like the way to work with a model the way the model understands you the way to talk with the model is kind of different thing than basically than the capabilities of the model I maybe May and part of June that model was Sol 5.6 I really loved working with that model and I kept working on that but ever since Opus 5.5 release even initial fable yes that was a leap in terms of intelligence but it wasn't so pleasant to work with and like right now Open AI is in trouble and because it was back and forth this year Open AI Entropic but ever since 5.5 it's been Entropic Entropic and Entropic I remember that quite a few episodes ago you unsubscribed from Cloud Code and basically doubled down on Open AI and now it's opposite yeah yeah exactly so it's kind of hard to be in that like in this kind of business because there is no loyalty in my opinion for the models yeah and so you either the best or well like don't get me wrong like you can most likely build everything with both models that you can think of and there won't be too much gap like the cloud models have a little bit better taste in terms of design that that's that's for sure but otherwise it's like no it's not like it's rocket science you know but what are you using for our podcast episode thumbnails is it cloud or is it chugpt both actually I am because I see better results when I tested some image generation and so on open AI models are somehow better I don't know cloud is struggling to generate kind of good thumbnails or images it's really hard to get something usable I'm not saying that chat GPT examples are great or perfect they are both bad but open AI is better yes yeah in terms of image generation of course yes open AI is way better I didn't even know that cloud and entropy can generate images they're trying but it's not good I said both because I've noticed a couple months ago that or well whenever the opus 5.5 I put that model in charge of the editing of the episode because previously I was using codex to do that but it's also in charge of generating the thumbnail but it figures out what's the best angle to generate the thumbnail and it uses OpenAI model to generate the image itself but it prompts it I made it so a cloud prompts OpenAI to generate the image you know so that I learned that this is the best approach for it had better results than me directly prompting OpenAI to generate those funnels yeah so there are use cases where you can find how to use both models and yeah but yeah I'm the same like you I probably could drop down my chat GPT subscription to $20 and just to see if I am hitting any limits probably I'm overpaying staying on $100 plan and I don't think we should be somehow loyal to one of those corporations which are making money of us use whatever is the best model which do the job and is most effective from revenue kind of basically cost effective and that's it yeah there's no reason to be loyal to any of those corporations just like being loyal to Apple basically do whatever it takes to say that product A is better than competitor than product B just because I'm a fan of it yeah AI models is the same consistently involving and the best model is always another one you know you can just and you can just switch it's nice that we can basically just jump in between them and keep using the best one yeah I agree and have you tried T3 code or not really we talked about it last week yes yes I tried it I set it basically usage okay if I'm building whatever Android application that's going on PC and I don't need to be like human like meet proxy which will kind of decide in which kind of like machine terminal to type that command I also I also building iOS applications but that's more for experiment on Linux it works but doesn't make sense if you have Mac around there is no benefits you can offload a little bit of work but it doesn't matter in my case but yeah it works but I have to say there's always a but yeah but the but is basically okay but why do I need it actually you for those who don't know that's basically like a layer on top of whatever AI models you have you have cloud you have codex and you can use T3 and this T3 is like another kind of abstraction level on top of your AI models and you are talking to your AI models through this T3 and it is able to create new subtasks and basically assign work to different models and the one user interface like cloud like open AI kind of models and cloud models and maybe grok what else not you can use all of them in this kind of T3 interface and then your kind of leading AI model will assign tasks to whoever it is needed but as we just discussed both of us are mostly using cloud for everything and you can basically use cloud application and use cloud for everything and for those kind of minor tasks when you need let's say codex to review pull request cloud can basically do that cloud can ask codex to do this code review get feedback work on pull request together that works okay we have this multi-machine use case well actually that works with cloud application as well I'm just running cloud application on both machines and basically I have two or three ways two ways maybe three whatever basically there is configuration options then you can basically kind of give a task from your Mac to execute on your PC which is next to you on desktop maybe using tilescale or what else not connection to connect remotely so I don't know I have it I have this T3 kind of setup and it kind of works but I have hard time to convince myself that I really need to use it well the reason I brought it up because I think it's the best right now experience you can get when working with any AI model yes I agree that we are using cloud but I won't be mad if OpenAI release a new model that's better to work with and I'll actually use it with and switch like like like I said there is no real loyalty either way like I think that's no loyalty to like any particular OpenAI model is actually good for tools like T3 code because like in my opinion it's the best experience you can get from mobile app tools like T3 code like gives us like that freedom you can just switch model and have the same experience in my opinion it's very valuable cool that's why I really like Hermes agent and you mentioned today dots by OpenAI I think like Grootbot is does that virtual assistant thing the best at the moment and Hermes is really great but it requires a little bit of tinkering and right now I'm using Grootbot for like tons of automation stuff etc and like what I was trying to do with Hermes was some kind of cron jobs automating stuff etc but I was missing that mobile app part of it and I built my own app to like have it like to manage it but I don't like maintaining mobile app for it that's why I'm kind of inclined a little bit to like I don't have like this kind of virtual assistant on Hermes but I for me and grogbot is really great for doing like computer use and for like tasks and in my opinion the integration with X that grogbot has is actually like for our show is really good because whatever is happening in like in grogbot doing like scouting it observes your social media my social media especially on X I don't have it like follow on other platforms which is an addition and I'm planning to do that but it builds the episode based of what we will talk about we can talk about I think it's really great tool and right now that's what I'm using for yeah I think you know that T3 is kind of giving you option to kind of plug in different AI models but from getting AI agnostic you are not tools agnostic anymore you are basically kind programming tasks but when you're doing any kind of research computer use tasks all that stuff what else we are doing then what are you still staying in T3 or you're switching back to cloud application and then how you manage jumping in between tools and kind of carrying over kind of whatever knowledge or memory what you did on one or another tool so there are a few things like before like getting into T3 code I switched for my coding tasks for from codex to cloud and that took some time but I switched although I still have codex open to like remember about some like coding tasks I did but as we said earlier I still use codex for like some computer use it's I don't switch I don't use T3 code for computer use specifically at the moment but that's what I said I actually like this separation having T3 code to actually be the tool for coding I really like this separation I I having grog bot to doing virtual assistant type of stuff and I prefer it this way because I know I am a very messy person like I am doing one thing here one thing here but I really like I am a messy person but I like structure it's kind of counterintuitive but that's who I am and I really like T3 code for many coding tasks and I'm scheduling some for example automations to for example I have like a few cron jobs running in cloud like one of it is automating automated CEO that runs every three days for that writes article for my one only one right now of my websites and to like generate some like views and like for me the most important part was and you can do it with tail scale of course and I think I have like the connection together with my MacBook and my Linux machine through T3 code is done with tail scale and you need that you can do that without T3 code but I don't know it feels so easy in my opinion to have this both those coding stuff done inside one app and managing switching like this part done on MacBook this part done on Linux machine inside one app it's kind of easy to manage those tools and especially I really like the mobile app yeah mobile app is good and it's I will say it's better for managing remote tasks than cloud application itself yeah makes sense for me it's kind of I set everything up started to use but then I was starting to think what's the point I'm kind of doing the problem is that lately I'm doing kind of less programming than kind of more thinkering around like kind of business analytics kind of generating some images talking about trying to brainstorm next move for let's say like like this and release it so before that I did some analytics kind of looking okay how much users are basically called seeds and so on you know and by doing some analytics I'm basically just asking cloud to pull the data because I have access to App Store Connect and I have access to analytic tools that I'm using in the application and so on and then okay where do I do that cloud application nice but when I the next step is okay go and implement so should I go now to the T3 and for me that kind of feels like I don't know and sure I can do like that in separate tasks in T3 as well but again it's like cloud has this cloud design kind of skill and you can use browser it then I think T3 is kind of great tool if you are more like product manager who also kind of do a little bit of programming which I am currently in my current state I basically kind of almost building nothing now and just kind of trying to figure out what is the next move for whatever let's say business what should I do next and so on looking into data analytics planning should I build new application or should I double down on any applications that I have in the app and so on that part somehow I don't know probably you can do all of that in you mean by saying you don't know which tool to use because I I've had very similar issue and yes exactly like T3 code is like as I said mainly for coding but all this business kind of stuff I still do right now inside T3 code but like I want to I like Grogbot like I mentioned for the virtual assistant stuff but I like thinking about I can get off the Grogbot subscription and being able to move this kind of virtual assistant stuff to my own solution or like based on Hermes or not only Hermes or maybe like now I started thinking about okay I'm most likely going to just make the wrapper and you can use you can use it with cloth or Hermes or Grock but that's what I like tinkering to do but yeah we'll see how this thing moves forward because tinkering and changing all time code of my own custom solution to myself is not going to bring me more clients to do marketing and yeah that's that's what I have to be aware of yeah I think we're going to finish the episode right now I have one more thing that do we need to do the follow up to previous episode I think we've done a segment last week on the 100k hackathon done by Mark Liu and Jack Freaks that's sponsored by Hicksfield I think just I think when I released the episode I kind of released the episode pretty fast after recording it Mark Liu like has been like there there has been some backlash that he got on X because of the entry fee basically for anybody that didn't listen to the episode Mark Liu is an indie hacker very popular in our scene and he decided to do a big hackathon with Hicksfield etc but he set up a fee to entry to the hackathon to $99 and he received tons of backlash on X about it that's what happened last week but just after we finished recording the episode last week he decided he said sorry he made a mistake and decided that the episode the hackathon will be for free that's the situation yeah basically as we expected that probably he will change it yeah and that happened even faster than we expected yeah yeah and I think the hackathon started yeah right I don't know where 30 days on 16th so it started two days ago on Monday okay yeah yeah so it started on Monday and now you can enter to win 100k for free I think it's a good decision I don't believe Mark the reason that he gave for the entry fee after recording the episode I read a little bit about it I think the reason for the entry fee that he gave is valid because yes you whenever you pay for something you actually like value it that's definitely a psychological thing that happens but I don't believe him that it was the it was the main thing because if that was the thing why increase the price after some time you know so yeah I saw someone pointed out some time ago like last week so that's why I said I don't believe that reason I think that's a good kind of good of cope out but not really what happened but I think it's good he admitted to being in mistake I don't know I'm not I'm not participating are you I'm not participating either I don't have plans I bet somebody will make tons of money and will blow out the competition and will just get with some kind of I don't know I bet it will be some kind because it's a business it's not like a SaaS or something I don't think that's the rules you just need to start a business to generate the most amount of revenue and if you are winning these competitions and you're probably already making great amount of money and you are winning basically double because you're getting prize money and able to launch successful good money generating business in 30 days so I do not participate those rules are limiting you quite a lot you know because we usually kind of building like subscription applications right yeah yeah and subscription applications that's usually kind of your focus is monthly recurring revenue and that grows over time you know yeah but here with this competition you basically have 30 days and you have ever built so that could be for example like I mentioned before like okay I'm running this 21k I would like to see some kind of application which is coach for 21k running and that could be you know something like it's like short period of time when you basically need to this program and some coaching you know to train for this kind of run one two three months maybe and then you probably could buy one time purchase to get some kind of program support and what else not you know that type of business that could be you know but also you need kind of scale you need something that everyone kind of happy to use that should be something way more bigger niche than somewhere those who are planning to run 21 kilometer for no good reason and even pay money to participate in that run so that's not good target audience you need something wider yeah yeah I have an idea for something like I said it's not an app it's a service basically kind of and I've been tinkering about this idea for a while I just thought of another angle to it but I don't think I'm going to start doing it either way but I don't think I'm going to apply to the hackathon because I I'm not 100% sure if I can fit within this timeline to generate the most amount of revenue in 30 days but we'll see I'm not sure if that is going to be a part of anything but I wish good luck to everyone who's participating I think I saw initial numbers they already have like I don't know about 3000 people already has signed up so yeah the competition is very steep yeah and let the best guy win yeah and that's it about today's episode I think I talked way too much today we need to arrange some topics that so Ivers would talk a little bit more than me I think my voice is a bit hurting right now but yeah thank you for watching the Fired By Robots podcast leave a comment subscribe we read all the comments I don't think we've got any to the latest episode so thank you once again for watching and see you in the next one cheers see you next week bye bye fire by robots podcast